Scene matching based on directional keylines and polar transform

Scene matching under complex background is a priority and difficulty in the field of computer vision, commonly used in matching real-time collected images and photos for navigation. Scene matching techniques are faced with complex natural scenes, the image distortion exist, such as light, change of gray levels, perspective, scaling and other differences, which cause matching difficult. This paper aims to find a scene matching algorithm; it has the characteristics of rotation and scaling invariance, anti-light and anti-slight-distortion, applicable for complex scene matching. The project has a new idea: combining the keylines with the vectors description based on polar image translation, and utilize the rotation-scale-invariance vectors to describe the extracted keylines, this method includes three steps: keylines extraction, description and matching. Preliminary experiments show that this keylines-based scene matching algorithm is applicable for image matching under complex background.